Failed-ML

Why we included this project

Most machine learning reading lists celebrate wins; this one collects the failures instead, and there is arguably more to learn from them. Each entry links to the original reporting, so you can read how Amazon's recruitment tool drifted into gender bias, why the COVID-19 triage models never reached clinical use, and where data leakage quietly invalidated published results. For engineers and data scientists, the value is pattern recognition: biased training data, evaluation shortcuts, and models that look fine in the lab and fall apart in production. The entries are grouped by domain, which makes it easy to find the failure modes closest to your own work. It is a reference to consult while designing a system, not a tool to run, and it works well as the starting point for a team post-mortem.